Note by @fastfinge

@FreakyFwoof @amir Okay! The problem is with how we clean accents. Not calling this function fixes it:

# --- HELPER: Strip accents to ensure pure ASCII ---
def clean_key_text(text):
# Normalize unicode to decompose characters (NFD)
# Then filter out non-spacing mark characters (accents)
return "".join(
c
for c in unicodedata.normalize("NFD", text)
if unicodedata.category(c) != "Mn"
)

But why? Are accents actually allowed in eloquence dictionaries? I didn't think so?